Converting a Nested 2D Loop to 1D#
On GPUs, a single 1D loop can be more efficient than a nested 2D loop for tasks like sampling a 2D window. Here’s how to sample a 3x3 window of offsets using a 1D loop.
Source Code#
// Window size.
const int WindowSize = 3;
const int WindowHalf = WindowSize / 2; // Integer division.
// 1D loop to iterate over 3x3 window.
for (int i = 0; i < (WindowSize * WindowSize); i++)
{
float2 XY = float2(i % WindowSize, i / WindowSize) - WindowHalf;
}
Example: 3x3 Window Offsets#
i# |
X |
Y |
||
---|---|---|---|---|
0 |
-1 + (0 % 3) |
-1 |
-1 + trunc(0 / 3) |
-1 |
1 |
-1 + (1 % 3) |
0 |
-1 + trunc(1 / 3) |
-1 |
2 |
-1 + (2 % 3) |
1 |
-1 + trunc(2 / 3) |
-1 |
3 |
-1 + (3 % 3) |
-1 |
-1 + trunc(3 / 3) |
0 |
4 |
-1 + (4 % 3) |
0 |
-1 + trunc(4 / 3) |
0 |
5 |
-1 + (5 % 3) |
1 |
-1 + trunc(5 / 3) |
0 |
6 |
-1 + (6 % 3) |
-1 |
-1 + trunc(6 / 3) |
1 |
7 |
-1 + (7 % 3) |
0 |
-1 + trunc(7 / 3) |
1 |
8 |
-1 + (8 % 3) |
1 |
-1 + trunc(8 / 3) |
1 |